A South Korean research team from the Korea Advanced Institute of Science and Technology (KAIST) has developed a quadrupedal robot that can navigate stairs and rough terrain without the use of visual or tactile sensors in a disaster situation where smoke and dust may obstruct its surroundings.
Named DreamWaQ, the robot control technology enables “blind locomotion” and takes its name from a situation in which a person can walk to the bathroom in the dark without visual assistance. Unlike traditional robot control methods that rely on kinematic or dynamic models, DreamWaQ employs a deep reinforcement learning approach that quickly computes the appropriate control commands for each motor of the robot based on various environmental data obtained from simulators.
The team said the new control system does not require separate tuning and can be easily applied to different types of walking robots. The learning process only takes about an hour, and the resulting behavior network is embedded in the actual robot. The robot can adapt to different environments by imagining similar scenarios to those it has experienced in simulators and using inertial sensors and joint angle measurements to move quickly and steadily.
DreamWaQ has successfully navigated environments such as university campuses with speed bumps and traffic calming measures, as well as outdoor areas with lots of tree roots and gravel, climbing stairs that are up to two-thirds the height of the robot’s body. It is capable of stable locomotion at speeds ranging from 0.3m/s to 1.0m/s, regardless of the environment.
The research results have been accepted for presentation at the IEEE International Conference on Robotics and Automation (ICRA) in London in late May.